PollenDetect

PollenDetect automates detection and classification of viable and nonviable pollen grains in microscopy images to enable high-throughput pollen viability assessment.

Key Features:

• Rapid detection: Reduces processing time from approximately 3 minutes per image to about 1 second, enabling high-throughput analysis.
• High accuracy: Achieves up to 99% accuracy in tests on cotton pollen viability and is comparable to 2,3,5-triphenyltetrazolium formazan quantification.
• Small-target-optimized detection: Implements the YOLOv5 neural network architecture adjusted for small target detection to effectively identify pollen grains.
• Viability classification: Distinguishes between viable and nonviable pollen grains for quantitative viability assessment.
• Trainability and adaptability: Can be further trained and customized to detect different types of pollen grains.

Scientific Applications:

• Agricultural screening: Enables rapid screening of stress-resistant crop varieties by assessing pollen viability under different conditions.
• Genetic breeding and gene identification: Assists in identifying key genes associated with pollen viability and stress resistance during genetic breeding programs.
• Environmental impact studies: Facilitates correlation of environmental factors, such as high temperatures, with reduced pollen viability to study climate change impacts on plant reproduction.

Methodology:

Uses deep learning via the YOLOv5 neural network model fine-tuned for small-target detection to detect and classify viable versus nonviable pollen grains.